diff --git a/sub_crates/backend/src/buffer.rs b/sub_crates/backend/src/buffer.rs
index 496ddf1..fe80ab5 100644
--- a/sub_crates/backend/src/buffer.rs
+++ b/sub_crates/backend/src/buffer.rs
@@ -1,21 +1,13 @@
 use ropey::Rope;
 
+use crate::marks::MarkSet;
+
 /// An open text buffer, currently being edited.
 #[derive(Debug, Clone)]
 pub struct Buffer {
-    pub is_dirty: bool, // Is this buffer currently out of sync with disk.
-    pub text: Rope,     // The actual text content.
-
-    // Sets of marked ranges to be used for various purposes.
-    //
-    // Each individual mark consists of a head and a tail.  The ordering of the
-    // head and tail is unspecified: either the head or the tail may come first
-    // in the text.  Both head and tail are specified in absolute indices.
-    //
-    // Within sets, marks cannot overlap or abutt, and they must be ordered.
-    // Overlapping or abutting marks will be merged, and out-of-order marks
-    // will be removed.
-    pub mark_sets: Vec<Vec<(usize, usize)>>,
+    pub is_dirty: bool,          // Is this buffer currently out of sync with disk.
+    pub text: Rope,              // The actual text content.
+    pub mark_sets: Vec<MarkSet>, // MarkSets for cursors, view positions, etc.
 }
 
 impl Buffer {
@@ -51,15 +43,7 @@ impl Buffer {
 
     /// Creates a new empty mark set, and returns the set index.
     pub fn add_mark_set(&mut self) -> usize {
-        self.mark_sets.push(vec![(0, 0)]);
+        self.mark_sets.push(MarkSet::new());
         return self.mark_sets.len() - 1;
     }
-
-    pub fn insert_new_mark(&mut self, _set_idx: usize, _mark: (usize, usize)) {
-        todo!()
-    }
-
-    pub fn clear_mark_set(&mut self, set_idx: usize) {
-        self.mark_sets[set_idx].clear();
-    }
 }
diff --git a/sub_crates/backend/src/lib.rs b/sub_crates/backend/src/lib.rs
index 8a18d5c..39d8765 100644
--- a/sub_crates/backend/src/lib.rs
+++ b/sub_crates/backend/src/lib.rs
@@ -4,4 +4,5 @@ extern crate unicode_segmentation;
 pub mod buffer;
 pub mod editor;
 pub mod hash;
+pub mod marks;
 pub mod project;
diff --git a/sub_crates/backend/src/marks.rs b/sub_crates/backend/src/marks.rs
new file mode 100644
index 0000000..104650f
--- /dev/null
+++ b/sub_crates/backend/src/marks.rs
@@ -0,0 +1,175 @@
+/// A mark on a piece of text, useful for representing cursors, selections, and
+/// general positions within a piece of text.
+///
+/// Both the head and tail are specified in absolute positions from the start
+/// of the text, and can have any relative relationship to each other (i.e. the
+/// tail can be before, after, or at the same position as the head).
+///
+/// For cursors and selections, the `head` should be considered the part that is
+/// moved by the user when e.g. extending selections.
+///
+/// `hh_pos` represents a target visual horizontal position of the mark's head,
+/// and is useful when e.g. moving cursors up/down vertically in the text.  But
+/// it can be ignored most of the time, does not effect editing operations in
+/// any way, and is frequently set to `None`.
+#[derive(Debug, Copy, Clone)]
+pub struct Mark {
+    pub head: usize,
+    pub tail: usize,
+    pub hh_pos: Option<usize>,
+}
+
+impl Mark {
+    pub fn new(head: usize, tail: usize) -> Mark {
+        Mark {
+            head: head,
+            tail: tail,
+            hh_pos: None,
+        }
+    }
+
+    /// Returns the properly sorted range of the mark.
+    pub fn range(&self) -> std::ops::Range<usize> {
+        if self.head < self.tail {
+            std::ops::Range::<usize> {
+                start: self.head,
+                end: self.tail,
+            }
+        } else {
+            std::ops::Range::<usize> {
+                start: self.tail,
+                end: self.head,
+            }
+        }
+    }
+
+    pub fn merge(&self, other: Mark) -> Mark {
+        let r1 = self.range();
+        let r2 = other.range();
+
+        let r3 = (r1.start.min(r2.start), r1.end.max(r2.end));
+
+        if self.head < self.tail {
+            Mark {
+                head: r3.0,
+                tail: r3.1,
+                hh_pos: None,
+            }
+        } else {
+            Mark {
+                head: r3.1,
+                tail: r3.0,
+                hh_pos: None,
+            }
+        }
+    }
+}
+
+//--------------------------------------------------------------------------
+
+/// A set of disjoint Marks, sorted by position in the text.
+///
+/// Because the `marks` `Vec` in this struct is publicly exposed, we can't
+/// actually guarantee that the marks are disjoint and sorted at all times, so
+/// do not rely on that for safety.  However, these invariants *should* hold,
+/// and code that modifies a MarkSet should ensure that the invariants remain
+/// true.
+///
+/// The `merge_touching` method will ensure that all expected invariants hold,
+/// modifying the set to meet the invariants if needed.
+#[derive(Debug, Clone)]
+pub struct MarkSet {
+    pub main_mark_idx: usize,
+    pub marks: Vec<Mark>,
+}
+
+impl MarkSet {
+    /// Creates an empty MarkSet.
+    pub fn new() -> MarkSet {
+        MarkSet {
+            main_mark_idx: 0,
+            marks: Vec::new(),
+        }
+    }
+
+    pub fn clear(&mut self) {
+        self.main_mark_idx = 0;
+        self.marks.clear();
+    }
+
+    /// Adds a new mark to the set, inserting it into its sorted position, and
+    /// returns the index where it was inserted.
+    ///
+    /// This assumes that all marks are already sorted by the start of their
+    /// range.
+    ///
+    /// This does *not* preserve disjointedness.  You should call
+    /// `merge_touching` after you have added all the marks you want.
+    ///
+    /// Runs in O(N + log N) time worst-case, but when the new mark is
+    /// inserted at the end of the set it is amortized O(1).
+    pub fn add_mark(&mut self, mark: Mark) -> usize {
+        // Special case optimization and early-out for inserting at the end of
+        // the set.
+        if self
+            .marks
+            .last()
+            .map(|l| l.range().start < mark.range().start)
+            .unwrap_or(true)
+        {
+            self.marks.push(mark);
+            return self.marks.len() - 1;
+        }
+
+        // Insert the mark.
+        let idx = self
+            .marks
+            .binary_search_by_key(&mark.range().start, |m| m.range().start)
+            .unwrap_or_else(|e| e);
+        self.marks.insert(idx, mark);
+
+        // Update the main_mark_idx.
+        if self.main_mark_idx >= idx && self.marks.len() > 1 {
+            self.main_mark_idx += 1;
+        }
+
+        idx
+    }
+
+    /// Merges all marks that are non-disjoint or not sorted relative to each
+    /// other.
+    ///
+    /// This results in a set of fully sorted, disjoint marks.  If the set
+    /// is already sorted and disjoint, no modifications are made.
+    ///
+    /// Note that even though the result is a sorted set, this _does not_ sort
+    /// the set in any expected way.  Rather, it merges marks that aren't
+    /// sorted.  For example, if the first mark is at the end of the text, and
+    /// the last mark is at the start of the text, all marks will be merged
+    /// into one.
+    ///
+    /// Runs in O(N) time.
+    pub fn merge_touching(&mut self) {
+        let mut i1 = 0;
+        let mut i2 = 1;
+
+        while i2 < self.marks.len() {
+            if self.marks[i1].range().end < self.marks[i2].range().start {
+                i1 += 1;
+                self.marks[i1] = self.marks[i2];
+                if self.main_mark_idx == i2 {
+                    self.main_mark_idx = i1;
+                }
+                i2 += 1;
+            } else {
+                self.marks[i1] = self.marks[i1].merge(self.marks[i2]);
+                if self.main_mark_idx == i2 {
+                    self.main_mark_idx = i1;
+                }
+                i2 += 1;
+            }
+        }
+
+        self.marks.truncate(i1 + 1);
+    }
+}